[Mono-dev] Use of mono-sgen causes mono path detection fallback

[Rodrigo Kumpera]

Good catch! I'm pushing a fix for it as of now, it will be part of mono
3.0.7.
Thanks for investigating it.

--
Rodrigo


On Mon, Mar 4, 2013 at 7:53 AM, Chris Ball <qhalid at gmail.com> wrote:

> I maintain a Linux build of mono (and F#) that is hosted via NFS, and used
> on a large number of machines, but the mount directory is not always
> identical. I have discovered that the mono-sgen binary does not properly
> detect the location of mscorlib on my systems, but the normal mono binary
> has no problems. This problem occurs in both 2.10.9 and 3.0.4. My "simple"
> test case (for mono 3.0) is to build the runtime, rename the output
> directory, and run the following:
>    MONO_GAC_PREFIX=/tmp/mono-test/renamed \
>   PATH=$PATH:/tmp/mono-test/renamed/bin \
>   LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/tmp/mono-test/renamed/lib \
>   /tmp/mono-test/renamed/bin/mono
> /tmp/mono-test/renamed/lib/mono/4.5/sn.exe
>    MONO_GAC_PREFIX=/tmp/mono-test/renamed \
>   PATH=$PATH:/tmp/mono-test/renamed/bin \
>   LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/tmp/mono-test/renamed/lib \
>   /tmp/mono-test/renamed/bin/mono-sgen
> /tmp/mono-test/renamed/lib/mono/4.5/sn.exe
>  The first command succeeds (prints out the usage). The second command
> fails, returning:
>    The assembly mscorlib.dll was not found or could not be loaded.
>   It should have been installed in the
> `/source/on/build/server/lib/mono/4.5/mscorlib.dll' directory.
> I have traced the issue to mono/metadata/assembly.c:compute_base:
>    /* Not a well known Mono executable, we are embedded, cant guess the
> base */
>   if (strcmp (p, "/mono") && strcmp (p, "/monodis") && strcmp (p, "/mint")
> && strcmp (p, "/monodiet"))
>     return NULL;
>
> By changing this as follows, the behavior is corrected:
>    /* Not a well known Mono executable, we are embedded, cant guess the
> base */
>   if (strcmp (p, "/mono") && strcmp (p, "/mono-sgen") && strcmp (p,
> "/monodis") && strcmp (p, "/mint") && strcmp (p, "/monodiet"))
>     return NULL;
> I have not filed this as a bug or submitted a patch yet, because I am not
> sure if other binaries should be included in this test as well (e.g.
> pedump). My preliminary tests using "pedump --verify all" suggest that it
> should also be added to the "well known" list.
>
> _______________________________________________
> Mono-devel-list mailing list
> Mono-devel-list at lists.ximian.com
> http://lists.ximian.com/mailman/listinfo/mono-devel-list
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ximian.com/pipermail/mono-devel-list/attachments/20130304/7cdc7934/attachment.html>